ElevenLabs Eleven v3
Audio Available Full comparison ↗ elevenlabs/eleven-v3 · by ElevenLabs
· neural-tts
Pricing — 1 offering(s)
About this offering: ElevenLabs' expressive text-to-speech model: Audio Tags such as [excited], [whispers] and [sighs] direct emotion and delivery, a dialogue mode handles multi-speaker scenes, and it covers 70+ languages. Generally available since 2026-02-02. ElevenLabs' newer Eleven v4 (released 2026-09-28) has the same $80 per 1M list price.
Text characters
- $100.00 / 1M characters
Historical
2026-07-30 → 2026-09-29
ElevenLabs' API pricing page listed Multilingual v2 and v3 as one row at $0.10 per 1K characters ($100 per…
ElevenLabs' API pricing page listed Multilingual v2 and v3 as one row at $0.10 per 1K characters ($100 per 1M).
- $80.00 / 1M characters
Current
2026-09-30 → present
List price on ElevenLabs' API pricing page: $0.08 per 1K characters ($80 per 1M), the same as Multilingual v2…
List price on ElevenLabs' API pricing page: $0.08 per 1K characters ($80 per 1M), the same as Multilingual v2 (checked 2026-10-11).
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how ElevenLabs Eleven v3 fits into a cost-aware routing setup
See how →Capability profile
How ElevenLabs Eleven v3 rates across core capability dimensions, with the task-level evidence behind each rating.
ElevenLabs' most expressive model — Audio Tags ([excited], [whispers], [sighs], etc.) give fine-grained emotional/delivery control beyond Multilingual v2's baseline. ElevenLabs' own GA testing reports a 68% error-rate reduction (15.3% to 4.9%) on numbers, symbols, and specialised notation across 8 languages vs the alpha release.
70+ languages per ElevenLabs' own v3 page — more than double Multilingual v2's 29.
Same 5,000+ voice library as Multilingual v2, plus a dialogue mode that weaves multiple voices into one multi-speaker generation, matching prosody and emotional range across speakers.
No latency figures published specific to v3 — not vendor-confirmed. Inferred comparable to Multilingual v2 (~400ms first-byte) as the shared base model; Audio Tags and dialogue-mode processing add overhead, so v3 is unlikely to be faster.
Full ElevenLabs voice cloning, same as Multilingual v2 and Flash v2.5.
Benchmarks
| Benchmark | Score | Config | Source |
|---|---|---|---|
| Pronunciation/notation accuracy improvement vs Eleven v3 alpha (vendor-reported) | 68 % | Error rate reduced from 15.3% to 4.9% across 27 categories (currency, chemical formulas, sports scores, phone numbers, etc.) in 8 languages, per ElevenLabs' own GA announcement. Not independently verified. | source ↗ |
Operator guidance
Choose Eleven v3 over Multilingual v2 when a script needs explicit emotional or delivery direction (Audio Tags) or multi-speaker dialogue, neither of which Multilingual v2 supports, at the same $80 per 1M characters. Eleven v4 (released 2026-09-28) now supersedes v3 at the same list price. Route latency-sensitive, real-time applications to Flash v2.5 instead: v3's audio-tag and dialogue features aren't built for low-latency streaming.
Use cases
- Expressive, emotionally-directed narration where Audio Tags control delivery
- Multi-speaker dialogue and conversational audio via dialogue mode
- Premium content where directed emotional performance matters more than low latency
Limitations
- No independently published latency figures — not confirmed suitable for real-time/streaming use
- Subscription plans bill in credits, so the effective rate depends on the plan; the price shown is the pay-as-you-go API rate
- Audio Tag behavior is 'somewhat voice and context dependent' per ElevenLabs' own documentation — not guaranteed uniform across all voices